New Benchmark Tests LLMs' Ability to Identify Code Errors in Competitive Programming
Researchers introduced UOJ-Bench, a benchmark that evaluates large language models on code generation, error detection, and code repair using real competitive programming submissions. Current LLMs fail to identify errors in over 50% of incorrect code submissions in standard evaluation, though performance improves to 90% with computationally expensive test-time scaling. The work suggests frontier LLMs could complement traditional online judge systems by catching errors humans miss, though practical deployment faces cost constraints.
A new research paper presents UOJ-Bench, a benchmark designed to assess how well large language models can support learning in competitive programming beyond basic problem-solving. The benchmark includes three tasks—code generation, code hacking (error identification), and code repair—all derived from real submissions on the Universal Online Judge platform. The study found that leading LLMs struggle significantly with error detection under standard conditions, succeeding in only about 50% of cases where code is known to be incorrect. However, when researchers applied test-time scaling techniques that increase computational resources, success rates climbed above 90%. Notably, the best-performing models with test-time scaling identified errors in over 5% of submissions that received perfect scores from standard judging systems, indicating potential value as a complementary tool. The authors acknowledge that the high computational costs of test-time scaling limit practical deployment at scale.
What's missing
The paper does not discuss potential limitations of using LLMs for code review in educational settings, such as whether students might over-rely on AI feedback or how this approach compares to human instructor feedback in terms of learning outcomes.
What different sources said
- arXiv cs.AICenter
Beyond Problem Solving: UOJ-Bench for Evaluating Code Generation, Hacking, and Repair in Competitive Programming
Related
Topology-Aware Thermodynamics Improves DNA Probe Specificity Design
Researchers developed a new framework for designing DNA probes that accounts for the spatial organization of matched sequences, not just overall thermodynamic stability. Traditional methods rely on scalar measures like melting temperature and free energy, which miss how mismatches are distributed along the probe. The approach could improve diagnostic accuracy in applications like HPV detection and gene expression profiling.
Study Identifies Optimal Thermal Dose for Combining Focused Ultrasound with Immunotherapy in Tumors
Researchers used multimodal PET imaging to identify an optimal thermal dose range for focused ultrasound ablation that destroys tumor tissue while preserving conditions for immunotherapy delivery. The study found that excessive heating collapses blood vessels needed for antibody access, while insufficient heating fails to adequately reduce tumor burden. The findings could guide clinical design of combination treatments pairing thermal ablation with immunotherapies.
Plant MSH1 Protein Functions as Mismatch-Directed Nuclease for Organelle Genome Maintenance
Researchers have identified the precise mechanism by which the AtMSH1 protein in Arabidopsis plants recognizes and cleaves DNA mismatches and lesions, preventing mutations in organellar genomes. The protein combines a DNA mismatch recognition module with a nuclease domain that makes staggered cuts at specific positions relative to DNA damage. This discovery explains how plants maintain unusually low mutation rates in their mitochondrial and chloroplast DNA compared to other eukaryotes.